gbm_slow_scoring.R

benchmark_scoring <- function(repetitions, rows, n.trees){
  # Score a n.trees gbm on rows, repetitions times, and time it.
  library("gbm")
  y        <- seq(1, 2, length.out = rows)
  number_x <- runif(rows)
  factor_x <- sample(letters, replace = TRUE, size = rows)
  factor_2 <- sample(LETTERS, replace = TRUE, size = rows)

  g <- gbm(y ~ number_x + factor_x,
           data = data.frame(y, number_x, factor_x),
           n.trees = n.trees)

  system.time(replicate(repetitions, predict(g, n.trees = n.trees)))
}

## Uncomment gbm source as appropriate
# devtools::install_github("gbm-developers/gbm")
# install.packages("gbm", repos = "http://cran.rstudio.com/")
benchmark_scoring(10, 1e4, 1e4)

# Takes ~ 85 seconds with github gbm and ~ 25 seconds with CRAN for me